Structure and ionic conduction enhancement mechanisms at CeO 2 /SrTiO 3 heterointerfaces

Fluorite-perovskite heterointerfaces garner great interest for enhanced ionic conductivity for application in electronic and energy devices. However, the origin of observed enhanced ionic conductivity as well as the details of the atomic structure at these interfaces remain elusive. Here, systematic, multi-stoichiometry computational searches and experimental investigations are performed to obtain stable and exact atomic structures of interfaces between CeO 2 and SrTiO 3 - two archetypes of the corresponding structural families. Local reconstructions take place at the interface because of mismatched lattices. TiO 2 terminated SrTiO 3 causes a buckled rock salt CeO interface layer to emerge. In contrast, SrO terminated SrTiO 3 maintains the fluorite structure at the interface compensated by a partially occupied anion lattice. Moderate enhancement in oxygen diffusion is found along the interface by simulations, yet evidence to support further significant enhancement is lacking. Our findings demonstrate the control of interface termination as an effective pathway to achieve desired device performance.


Introduction
Complex oxides exhibit a wide range of phenomena such as superconductivity, ionic conduction, magnetism and multi-ferroelectricity, and they have been intensively studied for applications that include solid oxide fuel cells (SOFCs), batteries, and memory devices 1 . 2In particular, oxide interfaces, like their conventional semiconductor counterparts, offer the unique opportunity to create tuneable novel multifunctionalities by the artificial interfacial modifications, including defects (oxygen or cation vacancies), structural symmetry breaking, interlayer interactions, etc. [3][4][5][6][7][8][9] For example, yttria-stabilized zirconia (YSZ) and doped CeO2 are the materials that have attracted intensive interest in SOFCs because of their chemical compatibility with electrodes and high ionic conductivity. 10,11In 2008, Garcia-Barriocanal et al. 12 reported an eight-orders-ofmagnitude enhancement of YSZ conductivity in epitaxial superlattices consisting of 1 nm YSZ and 10 nm SrTiO3 (STO) layers.The colossal enhancement of ionic conductivity was attributed to the combination of the large number of mobile ions together with the expansion of the fluorite structure imposed by the substrate. 13,14[20][21][22][23][24][25][26][27][28] Direct experimental measurements of interfaces are very challenging since interfaces have small volume fraction and are buried inside the sample.Theoretical and computational approaches can provide valuable insights into interface induced phenomena by idealized yet still realistic atomistic models.0][31] This is due to the difficulty in obtaining interface structures which are essential inputs for any modelling work.State-of-the-art scanning transmission electron microscopy (STEM) can reveal the locations of atomic columns at interfaces, but often cannot give a one-to-one mapping of the full lattice structure.This challenge can be tackled by performing direct computational searches for low energy interface structures.Indeed, recent advances in computational structure prediction algorithms [32][33][34][35] have led to discoveries of new high pressure materials 36 , defect clusters 37 , 2D materials 38 and interfaces. 39,40Combining both theoretical and experimental structure determination is necessary to get a complete understanding of the interface.
We focus on the CeO2/STO (001) interface because both CeO2 and STO are archetypes of their structural families and have technological importance.CeO2 has the cubic fluorite structure, and it can be tuned to become a high performance ionic conductor through doping with lower valence elements such as Sm and Gd. 41STO has a cubic perovskite structure and is commercially available as substrates for depositing epitaxial oxide thin films.The (001) CeO2/STO interfaces studied here can be realized by epitaxial growth of CeO2 thin films on (001)-oriented STO substrates.Previously, Dyer et al 29 studied the YSZ/STO interface using hand-built models and showed the importance of choosing surface terminations.Ultra-thin ZrO2/YSZ-STO superlattices have also been studied using interatomic potentials. 30,31Non-bulk phases were stabilized under both structural mismatch and strain.While YSZ also has the fluorite structure, its smaller lattice constants cause much larger tensile strain when epitaxially grown on STO (7% vs 2%), making it difficult to separate the effects of long ranged biaxial strain and localized interface structural mismatch.In addition, bulk YSZ itself may undergo phase transformation when subjected to strong biaxial tensile strain. 22,42 this work, stable and atomic interface structures at (001) CeO2/STO heterostructures are predicted using ab initio Random Structure Searching (AIRSS). 33,43A collection of low-energy interface structures has been obtained through systematic searching and has been further validated by STEM studies.The interface stoichiometry is found to play an important role in stabilizing the emergent atomic configurations, hence affecting the properties.With SrO terminated STO, the anion sites near the interface become partially occupied, giving rise to enhanced lateral oxygen diffusion up to two orders of magnitudes.On the other hand, structural mismatch at the TiO2 terminated interface is compensated by a buckled rock salt CeO.In both cases, additional oxygen vacancies are found to be attracted by the interface, which is expected to hinder ionic transport in the perpendicular direction.Overall, interface termination is shown to be an important factor producing interface-induced moderate ionic conductivity enhancement.

Finding interface structures
The bulk structures of STO and CeO2 are both cubic and they share some similarity of atomic arrangements (Figure 1).STO has a cubic perovskite structure of space group Pm3 ‾ m at room temperature while at low temperatures it turns into an anti-ferrodistortive tetragonal I4/mcm phase. 44We limit ourselves to the cubic phase as it is more relevant under growth and application conditions (from room temperature to hundreds of Kelvins).CeO2 has a cubic structure with Fm3 ̅ m space group.Cube-on-cube growth takes place when depositing CeO2 on a STO (001) surface, with the conventional cell of CeO2 rotated by 45 degrees along the z axis, 45 as depicted in Figure 1a.This gives the CeO2 an in-plane lattice constant of 3.82 Å, while that of STO is 3.905 Å.Hence, a CeO2 film is tensile strained by 2% when fully matched to the STO substrate.Such level of strain is reported to slightly enhance the oxygen diffusivity in the bulk structure. 20,46ormal to the interface.Similarly, STO is consisted of alternating SrO and TiO2 layers (Figure 1).Unlike CeO2, each layer in STO is charge neutral.Interestingly, the arrangement of oxygen anions in the TiO2 layer is the same as that in the O2 layer in CeO2.The cation sites occupied by Sr and Ti are translated by 0.5 ⃗ + 0.5 ⃗ ⃗ between each layers, similar to that of the Ce layers in CeO2.
Idealized interface models can be built by stacking blocks of the bulk structures in the [001] direction.There are four possible combinations: "SrO-Ce", "SrO-O2", "TiO2-Ce", "TiO2-O2".With the lateral alignments limited to three high symmetry arrangements, 12 interface structures can be enumerated.However, these models do not permit reconstructions that may lower the overall energy.In addition, the local stoichiometry at the interface is limited by the use of bulklike building blocks.Hence, as we will show below, performing variable composition structure searching is essential to acquire a complete picture of the interface and its properties.
Figure 2 shows interface structures obtained via first-principles structure searching.For SrO termination, there are three cases: SrO-Ce, SrO-O2 and SrO-O.The SrO-Ce case (Figure 2a) does not contain significant reconstruction, although a reduction of the Ce-O interface distance and buckling of the SrO layer can be seen.The lateral alignment of the layers is consistent with the minimization of the electrostatic energy.On the other hand, a significant reconstruction appears in the structure with the SrO-O2 termination.In Figure 2b, oxygen ions at the interface form O-O bonds with bond lengths between 1.45 Å and 1.50 Å which are similar to that of the peroxide groups.Three peroxide groups are in the SrO layer, and one is in the layer above.The formation of peroxide groups reduces the excessive negative charge at the interface due to O 2-.This also mitigates the polar discontinuity at the interface.The lowest energy structure found for the SrO-O type termination is shown in Figure 2c.Apart from the existence of vacant oxygen sites, shown as hollow red circles, the atomic arrangement in the interface layers appear to follow that of the bulk phase (as in Figure 1).There are three empty sites in the O2 layer and the other one is inside the SrO layer.We will come back to discuss the implications of these vacancies later.
Structures found for TiO2 terminated STO are shown in Figure 2d-f.The TiO2-Ce structure (Figure 2d) has little reconstruction at the interface.Also, the oxygen sub-lattice in the TiO2 layer is identical to that in bulk CeO2, which maintains 8-fold coordination environments of Ce atoms at the interface.The existence of Ti 4+ in the same layer, however, causes the inter-layer distance to increase through electrostatic repulsion.Similar to the SrO-O2 termination of Figure 2b, peroxide groups exist also in the structure with the TiO2-O2 termination, as shown in Figure 2e.The TiO2-O interface contains a buckled rock salt ordered CeO layer at the interface (Figure 2f).
The buckling is caused by the relatively small ionic radius of Ce 4+ as well as the repulsion from the O 2 layer above.

Terminations and stability
Relative stability of the interface models mentioned above can be compared via interface excess energy as a function of oxygen chemical potential (see SI for more details).In Figure 3, the difference in excess energy per unit area () is plotted against the oxygen chemical potential (  ′ =   −   0 ) for both SrO and TiO2 terminations, where   labels the relative oxygen contents.Only the lowest energy structures for each composition are included.At the oxygen deficient limit, i.e. when   ′ is highly negative, the SrO-Ce and TiO2-Ce (  = −4) configurations are stable.Similarly, the SrO-O2 and TiO2-O2 (  = 4) cases are stable at the oxygen rich limit.The SrO-O and TiO2-O interfaces are stable under intermediate condition and have   = 0 as indicated by the black lines in Figure 3.Other compositions with   between +4 and -4 are also included.Using the thermodynamic data of molecular oxygen, a phase diagram can be constructed (Figure S1).Both SrO-O and TiO2-O structures are found to be stable under typical film deposition conditions.Their structural parameters including the lattice constants and bond lengths are summarized in Table S3.The calculated projected density of states and band structures suggest the two interfaces remain electronically insulating (Figure S2-4 ).A shift of the conduction band minimum/valence band maximum can be seen in the layer-by-layer projected density of state, which can be related to the difference in the band gaps and electrostatic potential between SrTiO3 and CeO2.Semiconductor band bending at heterointerface is known to affect the electronic conductivity in other ionic conductor systems 47,48 .Interfaces with other terminations may become conductive as the Fermi level moves into the conduction band or the valence band as shown in Figure S5-6.

Vacancy formation energies
Having resolved the atomic structures at the interface, we now turn our focus to the properties they give rise to.The transport of oxygen ions in CeO2 relies on the existence of oxygen vacancies.Vacancy formation energies at various sites at the SrO-O (Figure 2c) and TiO2-O (Figure 2f) interface structures are shown in Figure 4.The lowest vacancy formation energy in each layer plotted against their  coordinates.Note that only the relative values are of the interest (Figure S7) here rather than the absolute vacancy formation energy which will require larger supercell to converge 49 .At the SrO-O interface, the formation energy is reduced from the bulk CeO2 value of 3.0 eV down to 2.3 eV.The decrease is more prominent in the TiO2-O structure, where the formation energy decreases to 1.5 eV at the interface and the reduction penetrates the CeO2 layers deeper.It was found that vacancies initially placed in the adjacent STO layer can automatically migrate to the interface during the geometry optimisation.

Ionic diffusivity
Ionic diffusivity can be strongly affected by the atomic structures and their corresponding energy landscapes.Unoccupied oxygen sites (i.e.oxygen vacancies) can potentially enhance diffusivity at the interface, which otherwise relies on intrinsic oxygen vacancies that are fewer in number in undoped material, giving poor ionic conductivities (~3.13 × 10 −3 S /cm at 1000 °C). 50We carried out molecular dynamics (MD) simulations to investigate oxygen diffusion at the interface based on the predicted the SrO-O interface structure (Figure 2c).Without explicitly introducing additional vacancies, the oxygen ions are found to be mobile in the lateral directions.Their trajectories within a period of 500 ps at 1600 K (Figure S8).Diffusion is found to be confined within two oxygen layers from the interface on each side.It is evident that such behaviour is strongly related to the interface reconstruction causing partially occupied oxygen sites.In addition, the lack of vacancy movement in the bulk region is consistent with the vacancy formation energy being reduced the interface as discussed above.On the other hand, no long-range diffusion at the TiO2-O interface has been found using the same simulation settings.
Oxygen diffusivity at 1000 K is estimated by extrapolating the fitted line in Figure S8, giving a value of 1.7 × 10 −8 cm 2 / s.2][53][54] Tensile strain is known to reduce the activation energy, and reductions of around ~0.05 eV based on theoretical studies have been reported for CeO2 about 2% tensile strain. 21,51Pristine CeO2 is expected to have low diffusivity (~1.0 × 10 −10 cm 2 / s) 55,56 due to the scarcity of oxygen vacancies as intrinsic defects, and here the interface gives rise to an enhancement of about two orders of magnitude.Repeating the MD simulations using Gd-doped CeO2 (10% at.) shows that the diffusivity at the interface region has the same order of magnitude compared to bulk GDC (Figure S9).This suggests that similar interfaces in doped/reduced CeO2 52,57 and YSZ 22 would have very limited effects on enhancing the overall oxygen diffusivity, as the bulk region is already highly ionically conductive.

Experimental verification
To verify the predicted interface structures, epitaxial thin films of CeO2 were grown on STO (001) substrates using pulsed-laser deposition and subsequently characterized under scanning transmission electron microscopy (STEM).Samples with both TiO2 and SrO termination were prepared, as described in the SI, where results of film quality and surface morphology characterizations are also displayed (Figure S13-S17).STEM is capable of imaging individual columns of atoms with multiple imaging modes.High-angle annular dark field (HAADF) images (Figure 5a&c) are used to reveal the positions of the metal cations, since it is based on "Zcontrast". 8,58The annular bright field (ABF) images (Figure 5b&d) are less Z-dependent, allowing the oxygen anion positions to be revealed, which are left out by the HAADF images.The experimentally observed interfaces match well with the predictions.In Figure 5a, the HAADF and elemental energy dispersive X-ray spectrometry (EDS) maps of the SrO terminated sample confirm the SrO termination, but the maps also indicate that the terminal SrO layer may exhibits some Ce/Sr intermixing (see SI for more).The ABF images for both SrO and TiO2 - terminated samples are shown in Figure 5b and Figure 5d respectively, with the corresponding SrO/TiO2-O interface models overlaid (Figure 2c&f).The assignment of layer identities is assisted with the HAADF and EDS data.In the ABF images of the SrO terminated sample, columns of atoms appear as dark dots, allowing the region of SrTiO3 and CeO2 to be identified.At the interface, an extra layer of dark dots (labeled "O" in Figure 5b) appears in a similar arrangement as the oxygen layer in CeO2, but its intensity is decreased.This feature is consistent with the prediction of a partially occupied oxygen layer in the SrO-O model.Although ABF cannot unambiguously differentiate between a fully occupied and a partially occupied O layer, our DFT calculations have shown that a fully occupied layer is not stable.The latter also results in oxygen anions moving off the lattice site (Figure 2b), which is not found in the ABF image, nor is any oxygen dumbbell structure.
For the TiO2 terminated interface, oxygen anions inside the terminal CeO layer can be clearly seen in the ABF image in Figure 5d (labeled "CeO") as smaller dark dots.In addition, they are displaced towards the STO side, which is consistent with the interface model in Figure 2f.Overall, the STEM images provide an invaluable support on the validity of our predictions of the interface structures and the calculated stabilities.The experimental results therefore underpin our calculation results on oxygen vacancy formation energy and diffusivity.

Discussion
Variation of the local composition has been suggested as the key to stabilizing interfaces 5,29,59 .Dyer et al. 29 studied the YSZ-STO interface by locally optimizing a range of hand-built models.It was reported that under the oxygen rich limit an interface containing a rock salt structured ZrO layer (  = 0) has lower excess energy compared to that with O2 (  = 4) or Zr (  = −4) termination.Our results here, obtained via structure searching without any a priori model, suggest that at the O-rich limit (  ′ = 0 eV) the stable structures contain peroxide groups (dumbbells, Figure 2b&e).The ionic interface reconstructions here can be seen as responding to both structural and chemical mismatches between the two materials, in analogy to the electronic reconstruction at the SrTiO3/LaAlO3 interface giving rise to a 2D electron gas.Modelling interfaces are computational challenging due to model size and complexities arise from local minima, choices of terminations, lateral alignment and systematic errors in the lattice parameters.Finding low energy configurations through structure searching is particularly useful, but the limit in model sizes and other approximations used still require careful convergence.Further works need to use large cell sizes with the help machine learning potentials which have gained significant improvements in recent years [60][61][62][63] .
Interfaces and grain boundaries are known to act as sinks of defects in ceramic materials.However, direct measurement of defect concentrations at interfaces is challenging and often limited to model systems.The effect of interfaces on vacancy stability has been reported in several theoretical works, but so far has been limited to interfaces between the same structure type 42 or has relied on hand-built models. 64,65The accumulation of vacancies at the STO-CeO2 interface may enhance the oxygen transport along the interface, but conduction is likely to be hindered perpendicular to the interface.This is because vacancies moving away from the interface requires additional energy.A space charge region would be present if the intrinsic concentration of the oxygen vacancies is low, i.e. when the CeO2 is undoped.Such a region is expected to extend much further into CeO2.A previous study on the Y2O3/CeO2 interface has estimated a Debye length of 3 nm, creating a zone of increased oxygen vacancy concentration around 10 nm which can enhance the ionic conductivity 66 .Our study shows that the interface terminations also play a role here.The TiO2-O terminated interface has a large reduction in the oxygen vacancy formation energy, which would result in stronger space charge effects.
The stable SrO-O interface structure (Figure 2c) contains many unoccupied oxygen sites (depicted by the empty circles).Dopants such as trivalent Sm or Gd, are known to induce charge compensation vacancies in CeO2, and thereby enhance the ionic diffusivity (and hence the conductivity).In analogy, a SrO termination layer may be seen as localized dopants which induces the vacancies.When half of the Ce 4+ atoms are replaced by Sr 2+ in CeO2, one in four oxygen sites should be vacant to balance the charge.Conversely, the lack of additional oxygen vacancies at a TiO2 terminated interface (Figure 2f) can be attributed to Ti and Ce having the same +4 oxidation state.Similar results of interface induced oxygen vacancies have been reported in CeO2/Y2O3 nanobrush structures where well-defined (111) interfaces are formed 67 .It was reported that more than 10% of the oxygen atoms are removed at the interface without inducing structural deterioration.
46,68 The estimated enhancement is limited to at most four-orders of magnitude (at 500 K) with a 4% tensile strain. 21In fact, excessive tensile strain may even reduce the diffusivity 46 and stabilize alternative phases of YSZ. 22,42,69Most of these studies have adopted bulk structure models rather than explicitly including the interface, neglecting the apparent strong mismatch in crystal structure between the fluorite/perovskite interface.Here, we fill the missing piece in by explicitly searching reconstructed interface structure that are low in energy.
Our results show that even with an emerging partially occupied oxygen sublattice localized at the interface, the magnitude of the resulting diffusivity is enhanced to a level similar to other doped fluorite conductors.Doped CeO2 already has many oxygen vacancies, and so any extra vacancies from the interface reconstruction are unlikely to give rise to further increases of the oxygen diffusivity.

Conclusion
In summary, we undertook a systematic search for stable STO/CeO2 interfaces using AIRSS.The obtained interface models were validated by the STEM images, and allowed the effect of interfaces on the ionic conductivity to be determined.Structural and chemical mismatching between STO and CeO2 was found to induce localized atomistic reconstructions, resulting in unoccupied oxygen sites when the STO was terminated by a SrO layer.MD simulations showed lateral oxygen ion diffusion localized at the interface, and the calculated diffusivity was found to be 1.7 × 10 −8 cm 2 / s at 1000 K, which is only a moderate enhancement from the bulk value, and is similar to that achieved in doped ceria.In fact, the interface itself acts as a region of localized aliovalent Sr 2+ doping.On the other hand, when STO is terminated by TiO2, the reconstructed interface contains a rock salt ordered CeO layer without any oxygen vacancies or any enhanced ionic conduction.For both terminations, additional oxygen vacancies are expected to be attracted to the interface which is detrimental to ionic transport perpendicular to the interface.While reconstructions can lead to locally enhanced ionic conductivity, which may have profound effects in nanoionics systems, its effect is limited when the materials are already bulk ionic conductors.Hence, we conclude the combined effect of interface structure and strain cannot lead to the eight orders of magnitude increase in ionic conductivity previously reported at YSZ/STO interface 12 .More broadly, our work shows that computational interface structure searching combined with STEM analysis provide an invaluable tool combination for modelling and understanding nontrivial heterogeneous interfaces, which are not possible to obtain otherwise due to the lack of reliable atomistic models.

Structure searching and DFT calculations
A detailed discussion about AIRSS, and its applications to interfaces can be found in the literature. 33,39,43,70,71Briefly, random but physically sensible interface structures are generated and relaxed to their nearby local minimum.The randomly generated structures are relaxed using standard DFT calculations without constraints and these relaxed structures are ranked by their energy.The process is repeated until sets of low energy structures are encountered multiple times.We impose bulk-like species-pair minimum distances constraints in the initial structure generation to limit the search space.Atoms within a single bulk (001) plane on each side of the interface are randomized.We also tested searches with an extended randomization zone up to two layers from each side but found no additional lower energy structures.The lateral alignment between the STO and CeO2 is varied during generation.No constraints are applied during local optimization, which allows the interface to drive the system into the optimum alignment.A slab cell construction with 15 Å vacuum is used for the initial search.The supercell contains 2 × 2 × 2 primitive cells of STO, with the bottom-most layer is fixed to the bulk coordinates.Two unit cells of bulk CeO2 are placed on top of the STO block.Half of the oxygen at the top CeO2(001) surface are removed according to experimentally observed surface reconstructions. 72o compare relative stability between different stoichiometries and avoid inclusion of surfaces the structures are converted into dual-interface models.This is possible because both STO and CeO2 have mirror planes parallel to (001) planes.
Density functional theory (DFT) calculations are performed using the plane-wave pseudopotential package CASTEP. 73Generalized gradient approximation (GGA) based exchange-correlation functionals have been used in many other computational studies of STO and CeO2 and are shown to give consistent results 42,[74][75][76][77] .We choose to use the PBEsol 78 exchange-correlation functional since it gives lattice parameters closer to the experimental values for both STO and CeO2.The valence states 2 2 2 4 for O, the 3 2 2 6 3 2 4 2 states for Ti, the 4 2 4 6 5 2 states for Sr and the 4 1 5 2 5 6 5 1 6 2 states for Ce are treated using on-the-fly generated core-corrected ultrasoft pseudopotentials from library QC5.A cut-off energy of 300 eV for the plane wave basis set is used in conjunction with soft pseudopotentials for initial searching, and a 2 × 2 × 1 k-point grid is used.Higher quality pseudopotentials (from library C9) are used for subsequent investigations with a cut-off energy of 600 eV and a 4 × 4 × 1 k-point grid.The one-the-fly generation strings of both sets of pseudopotentials are tabulated in the supplementary material.Since DFT is known to over-bind oxygen molecules (spin-triplet state), a 1.18 eV correction (per O2) is applied for the oxygen chemical potential reference, obtained by fitting experimental formation energies of alkali and alkaline metal oxides, as described in the literature 79 .Vacancy formation energies and transition state barrier calculation were performed using √2 × √2 supercells (2√2 × 2√2 of the perovskite primitive cell).We found this supercell size to be sufficient for obtaining relative differences in vacancy formation energies.The Atomic Simulation Environment 80 is used for model construction and conversion.DFT calculations are managed by AiiDA 81 which helps serving the provenance.The AIRSS code version 0.9.1 was used for generating random structures.

Molecular dynamic simulations
Buckingham potentials with long-range Coulomb interactions are used for molecular dynamics simulations performed with LAMMPS. 82Parameters of the potentials are described in the previous work. 70Oxygen mean squared displacements (MSDs) are extracted with respect to a range of time origins for averaging..The simulation box includes 8 × 8 unit cells in the x and y directions with a mirror slab geometry like the DFT calculations, and additional bulk layers have been inserted to increase the separation between the two mirror interfaces.The lattice constants are 31.0Å,31.0Å and 48.5Å along the x, y and z directions.The simulation includes a total of 3456 atoms and a time step of 2 fs.The NVT ensemble (Nosé -Hoover thermostat) is used with an initial temperature ramping stage of 20 ps.The simulation is then run for 500 ps to sample the MSD.The simulations were repeated for at least eight times for each temperature to provide adequate statistics.

Experimental
TiO2-terminated STO (001) substrates were used for growing CeO2 films from the chemically stoichiometric CeO2 ceramic targets by using a KrF excimer laser (248 nm) at a laser fluence of 2 J cm -2 .The growth temperature of 550 o C and oxygen partial pressure of 20 mTorr were used during the deposition.The termination of STO from TiO2 to SrO was achieved by using a SrRuO3 target to deposit an ultrathin layer monitored using reflective high energy electron diffraction (RHEED).Because the RuO2 layer is highly volatile, the film is self-terminated with the SrO surface afterwards. 83This process was carried out at 700 o C in 100 mTorr oxygen partial pressure and at a laser fluence of 1 J cm -2 .After growth, the films were cooled down to room temperature under an oxygen pressure of 20 mTorr.Electron microscopy analysis carried out afterwards confirms the absence of any residual RuO2, which otherwise may affect the interface properties. 84Film structures and surface morphologies were investigated by X-ray diffraction (XRD) on a high-resolution X-ray diffractometer (Empyrean, PANalytical, The Netherlands) using Cu Kα radiation (λ = 1.5405Å) and atomic force microscopy (AFM), respectively.Cross-sectional samples for STEM measurements were prepared using a focused ion beam (FIB) instrument.The high-angle annular dark field (HAADF) and annular bright field (ABF) images were acquired on a probe corrected STEM (FEI Titan Cubed Themis 60-300) operated at 300 kV.The collection angle range of the HAADF detector was 64~200 mrad, while the collection angle range of ABF images was calibrated at 8~53 mrad, acquired with a DF2 detector.

Figure 1 .
Figure 1.Bulk and interface strucutres.a) Bulk structures of CeO2 and STO in 3D.b) (001) planes in the CeO2 film layer, from the interface to the 3 rd layer.c) The (001) planes in the STO substrate.

Figure 3 .
Figure 3. Stability of the interfaces.Relative stability plot for (a) SrO and (b) TiO2 terminated interface structures with different oxygen stoichiometry.

Figure 4 .
Figure 4. Formation energy of oxygen vacancies.Oxygen vacancy formation energies plotted against  (parallel to the [001] direction) at the SrO-O and TiO2-O interfaces.Structures shown in Figure 2c&f are used for the computation.The interface is marked by the dotted line, and  = 0 corresponds to the location of the interface.

Figure 5 .
Figure 5. Scanning transmission electron microscopy images of the CeO2-STO interface in PLD-grown films of CeO2 on STO single crystal substrates, where the STO has a SrO (a, b) and a TiO2 (c, d) termination.The dark field images are shown side-by-side with the energy dispersive X-ray spectrometry (EDS) elemental maps in a) and c), which verify the designed terminations on the SrTiO3 side.In the ABF images b) and d), the predicted interface structures (Figure 2c&f), are overlayed on the annular bright field images.Color coding: Sr-green, Ti-blue, Ce-yellow, O-Red.